Claude took over the human computer for 12 hours: learned to slack off, and looked at the scenery while typing code

Hengyu from Aofei Temple
Quantum Bit | Public Account QbitAI

The new version of Claude 3.5 can use computers like humans, which makes us humans very excited!

After all, this means the beginning of a new competition: AI is no longer just about conversation and generation capabilities, but more about execution and operation.

In less than 12 hours, excited netizens have contributed their own videos of how they watched Claude play computer games.

In Anthropic’s announcement, there is another paragraph that aroused everyone’s interest:

…While recording the demonstration video, Claude accidentally stopped the screen recording program, resulting in the loss of all video footage.

Later, Claude took a break from a programming demonstration to look through photos of Yellowstone Park.

After all, it is expected that AI will make mistakes, but I don’t know where it learned that it needs to change its brain and take a rest after making mistakes.

There is video evidence, please watch the VCR: